- Gladstone AI report outlines action plan to prevent catastrophic AI risks, emphasizing national security measures and alignment with human intents.
- Report warns of new WMD-like risks from advanced AI, including weaponization and loss of control, proposing lines of effort to address these risks.
- Founders of Gladstone AI, Jeremie and Ed Harris, highlight the importance of early common-sense controls to ensure safe scaling of open-access AI models.
State Dept-backed report by Gladstone AI provides action plan to prevent catastrophic AI risks including weaponization and loss of control, emphasizing the need for national security measures and alignment between AI and human intents.
What to know: A report commissioned by the U.S. State Department suggests practical measures to prevent the emerging threats of advanced artificial intelligence, including the weaponization of AI and the threat of losing control over the technology. The report, titled, Defense in Depth: An Action Plan to Increase the Safety and Security of Advanced AI, was compiled by Gladstone AI, an AI safety company founded by brothers Jeremie and Edouard Harris. Work on the action plan began in October 2022, a month before the release of ChatGPT. It involved conversations with more than 200 people, including researchers and executives at frontier AI labs, cybersecurity experts, and national security officials in several countries.
Deeper details: The report warns that despite its immense benefits, advanced AI is creating entirely new categories of weapons of mass destruction-like (WMD-like) and WMD-enabling catastrophic risks. It highlights the risks of weaponization and loss of control. Weaponization includes risks such as AI systems autonomously discovering zero-day vulnerabilities, AI-powered disinformation campaigns, and bioweapon design. Loss of control suggests that as advanced AI approaches AGI-like levels of human- and superhuman general capability, it may become effectively uncontrollable, leading to power-seeking behaviors and deceptive manipulation of humans.
The backstory: The action plan introduced in the report focuses on lines of effort (LOE) to address the catastrophic national security risks of AI weaponization and loss of control without hindering the benefits of good AI use. The plan includes proposals to stabilize the current situation with respect to national security risks from AI RD, strengthen capabilities in AI safety and security, and establish legislative and international frameworks to scale up AI systems safely and securely. It also emphasizes the importance of AI safety research, establishing an AI regulatory agency, and diplomatic actions to establish an effective AI safeguards regime in international law.
Looking ahead: Jeremie and Ed Harris, founders of Gladstone AI, had concerns about AI risks since the release of GPT-2 in 2019 and became more earnest in their concerns with the release of GPT-3 in 2020. They recognized the scaling capabilities of AI and the potential risks associated with open-access models being fine-tuned for weaponization. Their experiences led them to exit their AI company and delve into addressing the risks associated with advanced AI. They emphasize the need for common-sense controls early on to ensure the safe and secure scaling of open-access AI models.
